一、目录
1 框架
2. 入门
3. 安装教程
4. 相关文档、案例阅读
二、实现
1 框架：Ray：将一个模型拆分到多个显卡中，实现分布式预测、训练等功能。
2. 入门 ：
案例：通过ray 实现分布式部署，分布式推理服务。
参考：https://zhuanlan.zhihu.com/p/647973148?utm_id=0
文件名：test.py
pip install ray
pip install “ray[serve]”
import pandas as pdimport ray
from ray import serve
from starlette.requests import Request@serve.deployment(ray_actor_options={"num_gpus": 2}) #两个gpu 将模型拆分，进行推理
class PredictDeployment:def __init__(self, model_id: str):from transformers import AutoModelForCausalLM, AutoTokenizerimport torchself.model = AutoModelForCausalLM.from_pretrained(model_id,torch_dtype=torch.float16,device_map="auto",)self.tokenizer = AutoTokenizer.from_pretrained(model_id)def generate(self, text: str) -> pd.DataFrame:input_ids = self.tokenizer(text, return_tensors="pt").input_ids.to(self.model.device)gen_tokens = self.model.generate(input_ids,temperature=0.9,max_length=200,)return pd.DataFrame(self.tokenizer.batch_decode(gen_tokens), columns=["responses"])async def __call__(self, http_request: Request) -> str: #异步实现http 框架json_request: str = await http_request.json()print(json_request)prompt=json_request[0]return self.generate(prompt["text"])deployment = PredictDeployment.bind(model_id="huggyllama/llama-13b")
运行>> serve run test:deployment #后台启动的 后台关闭指令： serve shutdown
测试：
import requests
sample_input = {"text": "Funniest joke ever:"}
output = requests.post("http://localhost:8000/", json=[sample_input]).json()
print(output)
- 安装教程
pip install ray
依赖环境：https://github.com/ray-project/ray - 相关文档、案例阅读
网址： https://github.com/ray-project/ray
文档-案例：https://docs.ray.io/en/latest/serve/index.html
